In the medical field, simulators such as operation simulators, organ simulators and the like are used to determine treatment plans, perform diagnosis, predict postoperative conditions, develop medical supplies and equipment and the like. In the simulation using these kinds of simulators, 3-dimensional shape data of an organ is used, however, often the generation of the 3-dimensional shape data of the organ is not easy. This is because the organs are located inside the body, so visual observation and direct measurement of the organs are not possible, and the shapes of the organs are very complex, fundamentally.
There is a conventional technique for generating a target shape by transforming a reference shape of an organ. Here, a method described below is known as a method for generating a target 3-dimensional shape data. Specifically, a doctor or the like observes tomographic images such as Computer Tomography (CT) images, Magnetic Resonance Imaging (MRI) images or the like, sets the boundaries of each portion of the organ, and draws boundary lines. Then, 3-dimensional shape data of an organ is obtained by laminating the tomographic images with the boundary lines.
However, it takes a long time to take one image in MRI or the like, sometimes sufficient tomographic images are not obtained (namely, a slice interval becomes long.) In this case, it is not possible to generate 3-dimensional shape data with high accuracy. For example, a shape that does not exist in a real organ is formed.
Moreover, there's a technique for transforming a reference model so as to optimize a predetermined evaluation function by using a model fitting method to generate a model of a target object. However, when resolution of a 3-dimensional image used for extracting the target object is low, 3-dimensional shape data with high accuracy is not obtained.
Moreover, there's a technique using a transformable model whose surface is formed by a mesh network. In this method, by repeatedly executing a step of newly calculating a position of a network point, 3-dimensional structured segmentation is generated from 2-dimensional images.
Moreover, there's a transformation method using a landmark as a parameter. In this transformation method, source landmarks are deployed on an area to be transformed, target landmarks are deployed on positions after transformation, and then transformation is executed. A method like this causes a problem that an unnatural shape would form unless the source landmarks and the target landmarks are set properly.
Furthermore, even though conventional techniques described above are used, accuracy of 3-dimensional shape data becomes low when sufficient tomographic images are not obtained.
Patent Document 1: Japanese Laid-open Patent Publication No. 2007-098028
Patent Document 2: Japanese Laid-open Patent Publication No. 2007-164592
Patent Document 3: Japanese Laid-open Patent Publication No. 2002-329216
Patent Document 4: Japanese Laid-open Patent Publication No. 2004-265337
Non-Patent Document 1: “Principal Warps: Thin-Plate Splines and the Decomposition of Deformations”, IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, Fred L. Bookstein, VOL. 11, NO. 6, June 1989
Non-Patent Document 2: “Laplacian surface editing”, SGP '04 Proceedings of the 2004 Eurographics/ACM SIGGRAPH symposium on Geometry processing, O. Sorkine, Tel Aviv University, D. Cohen-Or, Tel Aviv University, Y. Lipman, Tel Aviv University, M. Alexa, Darmstadt University of Technology, C. Roessl, Max-Planck Institut fuer Informatik, Saarbruecken, H.-P. Seidel, Max-Planck Institut fuer Informatik, Saarbruecken